We explore THz communication uplink multi-access with multi-hop Intelligent reflecting surfaces (IRSs) under correlated channels. Our aims are twofold: 1) enhancing the data rate of a desired user while dealing with interference from another user and 2) maximizing the combined data rate. Both tasks involve non-convex optimization challenges. For the first aim, we devise a sub-optimal analytical approach that focuses on maximizing the desired user’s received power, leading to an over-determined system. We also attempt to use approximate solutions utilizing pseudo-inverse (Pinv) and block solution (BLS) based methods. For the second aim, we establish a loose upper bound and employ an exhaustive search (ES). We employ deep reinforcement learning (DRL) to address both aims, demonstrating its effectiveness in complex scenarios. DRL outperforms mathematical approaches for the first aim, with the performance improvement of DDPG over the block solution ranging from 8% to 57.12%, and over the pseudo-inverse ranging from 41% to 190% for a correlation-factor equal to 1. Moreover, DRL closely approximates the ES for the second aim. Furthermore, our findings show that as channel correlation increases, DRL’s performance improves, capitalizing on the correlation for enhanced statistical learning.

Terahertz Multiple Access: A Deep Reinforcement Learning Controlled Multihop IRS Topology / Shehab, Muhammad; Elsayed, Mohamed; Almohamad, Abdullateef; Badawy, Ahmed; Khattab, Tamer; Zorba, Nizar; Hasna, Mazen; Trinchero, Daniele. - In: IEEE OPEN JOURNAL OF THE COMMUNICATIONS SOCIETY. - ISSN 2644-125X. - ELETTRONICO. - 5:(2024), pp. 1072-1087. [10.1109/OJCOMS.2024.3357701]

Terahertz Multiple Access: A Deep Reinforcement Learning Controlled Multihop IRS Topology

Shehab, Muhammad;Badawy, Ahmed;Trinchero, Daniele
2024

Abstract

We explore THz communication uplink multi-access with multi-hop Intelligent reflecting surfaces (IRSs) under correlated channels. Our aims are twofold: 1) enhancing the data rate of a desired user while dealing with interference from another user and 2) maximizing the combined data rate. Both tasks involve non-convex optimization challenges. For the first aim, we devise a sub-optimal analytical approach that focuses on maximizing the desired user’s received power, leading to an over-determined system. We also attempt to use approximate solutions utilizing pseudo-inverse (Pinv) and block solution (BLS) based methods. For the second aim, we establish a loose upper bound and employ an exhaustive search (ES). We employ deep reinforcement learning (DRL) to address both aims, demonstrating its effectiveness in complex scenarios. DRL outperforms mathematical approaches for the first aim, with the performance improvement of DDPG over the block solution ranging from 8% to 57.12%, and over the pseudo-inverse ranging from 41% to 190% for a correlation-factor equal to 1. Moreover, DRL closely approximates the ES for the second aim. Furthermore, our findings show that as channel correlation increases, DRL’s performance improves, capitalizing on the correlation for enhanced statistical learning.
File in questo prodotto:
File Dimensione Formato  
Trinchero-Terahertz.pdf

accesso aperto

Tipologia: 2. Post-print / Author's Accepted Manuscript
Licenza: Creative commons
Dimensione 1.15 MB
Formato Adobe PDF
1.15 MB Adobe PDF Visualizza/Apri
Trinchero-Terahertz_editoriale.pdf

accesso aperto

Tipologia: 2a Post-print versione editoriale / Version of Record
Licenza: Creative commons
Dimensione 2.89 MB
Formato Adobe PDF
2.89 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11583/2985361